Nomura Securities' robotics research report points out that the physical AI training data market is experiencing significant price differentiation. According to industry research, the price of first-person perspective video (Ego) data has dropped from $40-50 per hour in early 2026 to approximately $10-15 per hour or even lower currently, a decline of over 60%. With the entry of numerous suppliers, ordinary video data is gradually moving towards low-price competition, while teleoperation data, which includes real-time motion, contact, and force feedback, maintains a higher premium.
Nomura estimates that, based on the median hourly price of raw data, teleoperation data is more than seven times the price of Ego video; considering the actual usable time after filtering, the premium narrows to approximately 2-4 times.
This difference primarily stems from data quality: only 20%-30% of purchased Ego video passes screening, while the pass rate for teleoperation recording data is approximately 75%. Simultaneously, insufficient reliability and limited acquisition efficiency of haptic devices constrain the expansion of high-precision data supply.
Nomura believes that pricing power in the physical AI data industry is concentrating on high-quality contact data, data verification, and processing. Ordinary video capture providers are facing continued pressure to lower prices, while suppliers with the ability to capture realistic force and deliver high-quality data are expected to maintain their price advantage in the next procurement budget cycle.
